Zenkai Depth Prompt

Zenkai Depth Prompt is the depth-focused member of the Zenkai loader family. Where Zenkai Image Prompt V1 loads plain image+prompt pairs, this one loads depth maps with their matching prompts - the exact pairing you want for depth-conditioned generation (Depth ControlNet, depth-based relighting, or keeping composition stable across variations). One node, one folder, seed-chosen: IMAGE (the depth map) and STRING (the prompt) come out together.

It's a narrower tool than the Control Prompt nodes, and that's fine - a lot of people only need the depth condition, and this node skips the rest of the baggage.

How it works

Same folder pattern as the rest of the family, with its own directory: depthprompts/ in the pack install. The prompt_folder dropdown lists subfolders, auto-creating default on a fresh install. In a chosen subfolder, you place depth map files (jpg, jpeg, png) and a matching .txt for each - depth001.png pairs with depth001.txt, which should describe what the depth map depicts ("man holding umbrella, depth map of landscape," etc.).

Selection is the standard Zenkai behavior:

• mode: sequential - the seed walks the sorted folder index, looping; or random - seed-seeded, reproducible but varied.
• seed - the determinism knob. Same seed + same folder = same depth map and prompt.
• num_images (1–10) - how many depth maps to load in a batch.
• whitelist (optional) - comma-separated terms (phrases in quotes) that a prompt must contain for its depth map to be eligible. Empty = no filtering.

Outputs: IMAGE (the depth map(s)) and STRING (the prompt(s), joined with | when multiple). Wire the IMAGE into a depth ControlNet application node and the STRING into your text encoder.

Where it fits

If you've got a collection of depth maps you've generated or curated - say from a depth-estimation pass over a scene library - this node turns it into a searchable, seed-driven source of composition control. The KB's depth-estimation essay covers how depth maps are produced; this node is about the other half of the job, which is organizing and serving them.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/MushroomFleet/DJZ-Nodes
cd DJZ-Nodes
pip install -r requirements.txt

Restart ComfyUI, or search "DJZ-Nodes" in ComfyUI Manager. First run creates depthprompts/default/.

Troubleshooting

• Depth maps come out but prompts are empty. You've got image files without matching .txt files - unpaired depth maps are ignored or the prompt slot ends up empty. The pairing is name-for-name, so check base names match exactly.
• Only "default" in the dropdown. Fresh install. Create depthprompts/<your-category>/, drop files in, restart.
• Whitelist returns nothing. Substring matching strikes again - make sure your terms appear literally in the .txt prompts, and mind that man catches snowman.
• The "depth" looks flat/gray. That's usually the source material, not this node - it loads what's there. If you generated the depth with a weak model, the map will be shallow no matter how you serve it.

It's the specialist's choice in the family: one condition, one prompt, no extra wires. If depth is the only condition you use, this is cleaner than the full Control Prompt bundle - pick it accordingly.